James Slagle f619a0b59f Add docs for scaling for deployed-server
This documentation covers scaling up/down when using the set of
deployed-server templates.

Change-Id: I6f1420446a61546cf2b39dc671e2b01ad518c32d
2017-02-23 21:47:49 +00:00

14 KiB

Using Already Deployed Servers

TripleO can be used with servers that have already been deployed and provisioned with a running operating system.

In this deployment scenario, Nova and Ironic from the Undercloud are not used to do any server deployment, installation, or power management. An external to TripleO and already existing provisioning tool is expected to have already installed an operating system on the servers that are intended to be used as nodes in the Overcloud.

Note

It's an all or nothing approach when using already deployed servers. Mixing using deployed servers with servers provisioned with Nova and Ironic is not currently possible.

Deployed Server Requirements

Networking

Undercloud

Neutron in the Undercloud is not used for providing DHCP services for the Overcloud nodes, hence a dedicated provisioning network with L2 connectivity is not a requirement in this scenario. Neutron is however still used for IPAM for the purposes of assigning IP addresses to the port resources created by tripleo-heat-templates.

Network L3 connectivity is still a requirement between the Undercloud and Overcloud nodes. The Overcloud nodes will communicate over HTTP(s) to poll the Undercloud for software configuration to be applied by their local agents. Thus, the polling process requires that the Undercloud services are bound to an IP address that is on a L3 routed network that is accessible to the Overcloud nodes. This is the IP address that is configured via local_ip in the undercloud.conf file used during Undercloud installation.

On each Overcloud node run the following commands that test connectivity to the Undercloud's IP address where OpenStack services are bound.

Test basic connectivity to the Undercloud:

ping <undercloud local_ip>

Test HTTP/HTTPS connectivity to Heat API on the Undercloud:

curl <undercloud local_ip>:8000

Sample output:

{"versions": [{"status": "CURRENT", "id": "v1.0", "links": [{"href": "http://10.12.53.41:8000/v1/", "rel": "self"}]}]}

Test HTTP/HTTPS connectivity to Swift on the Undercloud The html output shown here is expected! While it indicates no resource was found, it demonstrates successful connectivity to the HTTP service:

curl <undercloud local_ip>:8080

Sample output:

<html><h1>Not Found</h1><p>The resource could not be found.</p></html>

The output from the above curl commands demonstrates successful connectivity to the web services bound at the Undercloud's local_ip IP address. It's important to verify this connectivity prior to starting the deployment, otherwise the deployment may be unsuccessful and difficult to debug.

Overcloud

Other than not having to be connected to a dedicated L2 network for provisioning by the Undercloud, networking requirements for deployed servers that will be used as nodes in the Overcloud remains unchanged.

Depending on what networking configuration and if network isolation is in use determines the requirements around Overcloud network requirements, the same as in a typical TripleO deployment where already deployed servers are not used.

Package repositories

The servers will need to already have the appropriately enabled yum repositories as packages will be installed on the servers during the Overcloud deployment. The enabling of repositories on the Overcloud nodes is the same as it is for other areas of TripleO, such as Undercloud installation. See ../basic_deployment/repositories for the detailed steps on how to enable the standard repositories for TripleO.

Initial Package Installation

Once the repositories have been enabled on the deployed servers, the initial packages for the Heat agent need to be installed. Run the following command on each server intending to be used as part of the Overcloud:

sudo yum -y install python-heat-agent*

Deploying the Overcloud

Deployment Command

The functionality of using already deployed servers is enabled by passing additional Heat environment files to the openstack overcloud deploy command.:

openstack overcloud deploy \
  <other cli arguments> \
  --disable-validations \
  -e /usr/share/openstack-tripleo-heat-templates/environments/deployed-server-environment.yaml \
  -e /usr/share/openstack-tripleo-heat-templates/environments/deployed-server-bootstrap-centos.yaml \
  -r /usr/share/openstack-tripleo-heat-templates/deployed-server/deployed-server-roles-data.yaml

The --disable-validations option disables the basic Nova, Ironic, and Glance related validations executed by python-tripleoclient. These validations are not necessary since those services will not be used to deploy the Overcloud.

The deployed-server.yaml environment takes advantage of the template composition nature of Heat and tripleo-heat-templates to substitute OS::Heat::DeployedServer resources in place of OS::Nova::Server.

The deployed-server-bootstrap-centos.yaml environment triggers execution of a bootstrap script on the deployed servers to install further needed packages and make other configurations necessary for Overcloud deployment.

The custom roles file, deployed-server-roles-data.yaml contains the custom roles used during the deployment. Further customization of the roles data is possible when using deployed servers. When doing so, be sure to include the disable_constraints key on each defined role as seen in deployed-server-roles-data.yaml. This key disables the Heat defined constraints in the generated role templates. These constraints validate resources such as Nova flavors and Glance images, resources that are not needed when using deployed servers. An example role using disable_constraints looks like:

- name: ControllerDeployedServer
  disable_constraints: True
  CountDefault: 1
  ServicesDefault:
    - OS::TripleO::Services::CACerts
    - OS::TripleO::Services::CephMon
    - OS::TripleO::Services::CephExternal
    - OS::TripleO::Services::CephRgw
    ... <additional services>

Configuring Deployed Servers to poll Heat

Upon executing the deployment command, Heat will begin creating the overcloud stack. The stack events are shown in the terminal as the stack operation is in progress.

The resources corresponding to the deployed server will enter CREATE_IN_PROGRESS. At this point, the Heat stack will not continue as it is waiting for signals from the servers. The agents on the deployed servers need to be configured to poll Heat for their configuration.

This point in the Heat events output will look similar to:

2017-01-14 13:25:13Z [overcloud.Compute.0.NovaCompute]: CREATE_IN_PROGRESS  state changed
2017-01-14 13:25:14Z [overcloud.Controller.0.Controller]: CREATE_IN_PROGRESS  state changed
2017-01-14 13:25:14Z [overcloud.Controller.1.Controller]: CREATE_IN_PROGRESS  state changed
2017-01-14 13:25:15Z [overcloud.Controller.2.Controller]: CREATE_IN_PROGRESS  state changed

The example output above is from a deployment with 3 controllers and 1 compute. As seen, these resources have entered the CREATE_IN_PROGRESS state.

To configure the agents on the deployed servers, the request metadata url needs to be read from Heat resource metadata on the individual resources, and configured in the /etc/os-collect-config.conf configuration file on the corresponding deployed servers.

Manual configuration of Heat agents

These steps can be used to manually configure the Heat agents (os-collect-config) on the deployed servers.

Query Heat for the request metadata url by first listing the nested deployed-server resources:

openstack stack resource list -n 5 overcloud | grep deployed-server

Example output:

| deployed-server | 895c08b8-f6f4-4564-b344-586603e7e970 | OS::Heat::DeployedServer | CREATE_COMPLETE    | 2017-01-14T13:25:12Z | overcloud-Controller-pgeu4nxsuq6r-1-v4slfaduprak-Controller-ltxdxz2fin3d |
| deployed-server | 87cd8d81-8bbe-4c0b-9bd9-f5bcd1343265 | OS::Heat::DeployedServer | CREATE_COMPLETE    | 2017-01-14T13:25:15Z | overcloud-Controller-pgeu4nxsuq6r-0-5uin56wp3ign-Controller-5wkislg4kiv5 |
| deployed-server | 3d387f61-dc6d-41f7-b3b8-5c9a0ab0ed7b | OS::Heat::DeployedServer | CREATE_COMPLETE    | 2017-01-14T13:25:16Z | overcloud-Controller-pgeu4nxsuq6r-2-m6tgzatgnqrb-Controller-yczqaulovrla |
| deployed-server | cc230478-287e-4591-a905-bbfca6c89742 | OS::Heat::DeployedServer | CREATE_COMPLETE    | 2017-01-14T13:25:13Z | overcloud-Compute-vllmnqf5d77h-0-kfm2xsdmtmr6-NovaCompute-67djxtyrwi6z |

Show the resource metadata for one of the resources. The last column in the above output is a nested stack name and is used in the command below. The command shows the resource metadata for the first controller (Controller.0):

openstack stack resource metadata overcloud-Controller-pgeu4nxsuq6r-0-5uin56wp3ign-Controller-5wkislg4kiv5 deployed-server

The above command outputs a significant amount of JSON output representing the resource metadata. To see just the request metadata_url, the command can be piped to jq to show just the needed url:

openstack stack resource metadata overcloud-Controller-pgeu4nxsuq6r-0-5uin56wp3ign-Controller-5wkislg4kiv5 deployed-server | jq -r '.["os-collect-config"].request.metadata_url'

Example output:

http://10.12.53.41:8080/v1/AUTH_cf85adf63bc04912854473ff2b08b5a2/ov-ntroller-5wkislg4kiv5-deployed-server-yc4lx2d43dmb/244744c2-4af1-4626-92c6-94b2f78e3791?temp_url_sig=6d33b16ee2ae166a306633f04376ee54f0451ae4&temp_url_expires=2147483586

Using the above url, configure /etc/os-collect-config.conf on the deployed server that is intended to be used as Controller 0. The full configuration would be:

[DEFAULT]
collectors=request
command=os-refresh-config
polling_interval=30

[request]
metadata_url=http://10.12.53.41:8080/v1/AUTH_cf85adf63bc04912854473ff2b08b5a2/ov-ntroller-5wkislg4kiv5-deployed-server-yc4lx2d43dmb/244744c2-4af1-4626-92c6-94b2f78e3791?temp_url_sig=6d33b16ee2ae166a306633f04376ee54f0451ae4&temp_url_expires=2147483586

Once the configuration has been updated on the deployed server for Controller 0, restart the os-collect-config service:

sudo systemctl restart os-collect-config

Repeat the configuration for the other nodes in the Overcloud, by querying Heat for the request metadata url, and updating the os-collect-config configuration on the respective deployed servers.

Once all the agents have been properly configured, they will begin polling for the software deployments to apply locally from Heat, and the Heat stack will continue creating. If the deployment is successful, the Heat stack will eventually go to the CREATE_COMPLETE state.

Automatic configuration of Heat agents

A script is included with tripleo-heat-templates that can be used to do automatic configuration of the Heat agent on the deployed servers instead of relying on the above manual process.

The script requires that the environment variables needed to authenticate with the Undercloud's keystone have been set in the current shell. These environment variables can be set by sourcing the Undercloud's stackrc file.

The script also requires that the user executing the script can ssh as the same user to each deployed server, and that the remote user account has password-less sudo access.

The following shows an example of running the script:

export OVERCLOUD_ROLES="ControllerDeployedServer ComputeDeployedServer"
export ControllerDeployedServer_hosts="192.168.25.1 192.168.25.2 192.168.25.3"
export ComputeDeployedServer_hosts="192.168.25.4"
tripleo-heat-templates/deployed-server/scripts/get-occ-config.sh

As shown above, the script is further configured by the $OVERCLOUD_ROLES environment variable, and corresponding $<role-name>_hosts variables.

$OVERCLOUD_ROLES is a space separated list of the role names used for the Overcloud deployment. These role names correspond to the name of the roles from the roles data file used during the deployment.

Each $<role-name>_hosts variable is a space separated list of IP addresses that are the IP addresses of the deployed servers for the roles. For example, in the above command, 192.168.25.1 is the IP of Controller 0, 192.168.25.2 is the IP of Controller 1, etc.

The script will take care of querying Heat for each request metadata url, configure the url in the agent configuration file on each deployed server, and restart the agent service.

Once the script executes successfully, the deployed servers will start polling Heat for software deployments and the stack will continue creating.

Scaling the Overcloud

Scaling Up

When scaling up the Overcloud, the heat agents on the new servers being added to the deployment need to be configured to correspond to the new nodes being added.

For example, when scaling out compute nodes, the steps to be completed by the user are as follows:

  1. Prepare the new deployed server(s) as shown in Deployed Server Requirements.
  2. Start the scale out command. See scale_roles for reference.
  3. Once Heat has created the new resources for the new deployed server(s), query Heat for the request metadata url for the new nodes, and configure the remote agents as shown in Manual configuration of Heat agents.

Scaling Down

When scaling down the Overcloud, follow the scale down instructions as normal as shown in delete_nodes.

The physical deployed servers that have been removed from the deployment need to be powered off. In a deployment not using deployed servers, this would typically be done with Ironic. When using deployed servers, it must be done manually, or by whatever existing power management solution is already in place. If the nodes are not powered down, they will continue to be operational and could be part of the deployment, since there are no steps to unconfigure, uninstall software, or stop services on nodes when scaling down.

Once the nodes are powered down and all needed data has been saved from the nodes, it is recommended that they be reprovisioned back to a base operating system configuration so that they do not unintentionally join the deployment in the future if they are powered back on.

Note

Do not attempt to reuse nodes that were previously removed from the deployment without first reprovisioning them using whatever provisioning tool is in place.